linux/fs/block_dev.c
<<
>>
Prefs
   1/*
   2 *  linux/fs/block_dev.c
   3 *
   4 *  Copyright (C) 1991, 1992  Linus Torvalds
   5 *  Copyright (C) 2001  Andrea Arcangeli <andrea@suse.de> SuSE
   6 */
   7
   8#include <linux/init.h>
   9#include <linux/mm.h>
  10#include <linux/fcntl.h>
  11#include <linux/slab.h>
  12#include <linux/kmod.h>
  13#include <linux/major.h>
  14#include <linux/device_cgroup.h>
  15#include <linux/highmem.h>
  16#include <linux/blkdev.h>
  17#include <linux/module.h>
  18#include <linux/blkpg.h>
  19#include <linux/magic.h>
  20#include <linux/buffer_head.h>
  21#include <linux/swap.h>
  22#include <linux/pagevec.h>
  23#include <linux/writeback.h>
  24#include <linux/mpage.h>
  25#include <linux/mount.h>
  26#include <linux/uio.h>
  27#include <linux/namei.h>
  28#include <linux/log2.h>
  29#include <linux/cleancache.h>
  30#include <linux/aio.h>
  31#include <asm/uaccess.h>
  32#include "internal.h"
  33
  34struct bdev_inode {
  35        struct block_device bdev;
  36        struct inode vfs_inode;
  37};
  38
  39static const struct address_space_operations def_blk_aops;
  40
  41static inline struct bdev_inode *BDEV_I(struct inode *inode)
  42{
  43        return container_of(inode, struct bdev_inode, vfs_inode);
  44}
  45
  46inline struct block_device *I_BDEV(struct inode *inode)
  47{
  48        return &BDEV_I(inode)->bdev;
  49}
  50EXPORT_SYMBOL(I_BDEV);
  51
  52/*
  53 * Move the inode from its current bdi to a new bdi. If the inode is dirty we
  54 * need to move it onto the dirty list of @dst so that the inode is always on
  55 * the right list.
  56 */
  57static void bdev_inode_switch_bdi(struct inode *inode,
  58                        struct backing_dev_info *dst)
  59{
  60        struct backing_dev_info *old = inode->i_data.backing_dev_info;
  61
  62        if (unlikely(dst == old))               /* deadlock avoidance */
  63                return;
  64        bdi_lock_two(&old->wb, &dst->wb);
  65        spin_lock(&inode->i_lock);
  66        inode->i_data.backing_dev_info = dst;
  67        if (inode->i_state & I_DIRTY)
  68                list_move(&inode->i_wb_list, &dst->wb.b_dirty);
  69        spin_unlock(&inode->i_lock);
  70        spin_unlock(&old->wb.list_lock);
  71        spin_unlock(&dst->wb.list_lock);
  72}
  73
  74/* Kill _all_ buffers and pagecache , dirty or not.. */
  75void kill_bdev(struct block_device *bdev)
  76{
  77        struct address_space *mapping = bdev->bd_inode->i_mapping;
  78
  79        if (mapping->nrpages == 0)
  80                return;
  81
  82        invalidate_bh_lrus();
  83        truncate_inode_pages(mapping, 0);
  84}       
  85EXPORT_SYMBOL(kill_bdev);
  86
  87/* Invalidate clean unused buffers and pagecache. */
  88void invalidate_bdev(struct block_device *bdev)
  89{
  90        struct address_space *mapping = bdev->bd_inode->i_mapping;
  91
  92        if (mapping->nrpages == 0)
  93                return;
  94
  95        invalidate_bh_lrus();
  96        lru_add_drain_all();    /* make sure all lru add caches are flushed */
  97        invalidate_mapping_pages(mapping, 0, -1);
  98        /* 99% of the time, we don't need to flush the cleancache on the bdev.
  99         * But, for the strange corners, lets be cautious
 100         */
 101        cleancache_invalidate_inode(mapping);
 102}
 103EXPORT_SYMBOL(invalidate_bdev);
 104
 105int set_blocksize(struct block_device *bdev, int size)
 106{
 107        /* Size must be a power of two, and between 512 and PAGE_SIZE */
 108        if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
 109                return -EINVAL;
 110
 111        /* Size cannot be smaller than the size supported by the device */
 112        if (size < bdev_logical_block_size(bdev))
 113                return -EINVAL;
 114
 115        /* Don't change the size if it is same as current */
 116        if (bdev->bd_block_size != size) {
 117                sync_blockdev(bdev);
 118                bdev->bd_block_size = size;
 119                bdev->bd_inode->i_blkbits = blksize_bits(size);
 120                kill_bdev(bdev);
 121        }
 122        return 0;
 123}
 124
 125EXPORT_SYMBOL(set_blocksize);
 126
 127int sb_set_blocksize(struct super_block *sb, int size)
 128{
 129        if (set_blocksize(sb->s_bdev, size))
 130                return 0;
 131        /* If we get here, we know size is power of two
 132         * and it's value is between 512 and PAGE_SIZE */
 133        sb->s_blocksize = size;
 134        sb->s_blocksize_bits = blksize_bits(size);
 135        return sb->s_blocksize;
 136}
 137
 138EXPORT_SYMBOL(sb_set_blocksize);
 139
 140int sb_min_blocksize(struct super_block *sb, int size)
 141{
 142        int minsize = bdev_logical_block_size(sb->s_bdev);
 143        if (size < minsize)
 144                size = minsize;
 145        return sb_set_blocksize(sb, size);
 146}
 147
 148EXPORT_SYMBOL(sb_min_blocksize);
 149
 150static int
 151blkdev_get_block(struct inode *inode, sector_t iblock,
 152                struct buffer_head *bh, int create)
 153{
 154        bh->b_bdev = I_BDEV(inode);
 155        bh->b_blocknr = iblock;
 156        set_buffer_mapped(bh);
 157        return 0;
 158}
 159
 160static ssize_t
 161blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 162                        loff_t offset, unsigned long nr_segs)
 163{
 164        struct file *file = iocb->ki_filp;
 165        struct inode *inode = file->f_mapping->host;
 166
 167        return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
 168                                    nr_segs, blkdev_get_block, NULL, NULL, 0);
 169}
 170
 171int __sync_blockdev(struct block_device *bdev, int wait)
 172{
 173        if (!bdev)
 174                return 0;
 175        if (!wait)
 176                return filemap_flush(bdev->bd_inode->i_mapping);
 177        return filemap_write_and_wait(bdev->bd_inode->i_mapping);
 178}
 179
 180/*
 181 * Write out and wait upon all the dirty data associated with a block
 182 * device via its mapping.  Does not take the superblock lock.
 183 */
 184int sync_blockdev(struct block_device *bdev)
 185{
 186        return __sync_blockdev(bdev, 1);
 187}
 188EXPORT_SYMBOL(sync_blockdev);
 189
 190/*
 191 * Write out and wait upon all dirty data associated with this
 192 * device.   Filesystem data as well as the underlying block
 193 * device.  Takes the superblock lock.
 194 */
 195int fsync_bdev(struct block_device *bdev)
 196{
 197        struct super_block *sb = get_super(bdev);
 198        if (sb) {
 199                int res = sync_filesystem(sb);
 200                drop_super(sb);
 201                return res;
 202        }
 203        return sync_blockdev(bdev);
 204}
 205EXPORT_SYMBOL(fsync_bdev);
 206
 207/**
 208 * freeze_bdev  --  lock a filesystem and force it into a consistent state
 209 * @bdev:       blockdevice to lock
 210 *
 211 * If a superblock is found on this device, we take the s_umount semaphore
 212 * on it to make sure nobody unmounts until the snapshot creation is done.
 213 * The reference counter (bd_fsfreeze_count) guarantees that only the last
 214 * unfreeze process can unfreeze the frozen filesystem actually when multiple
 215 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
 216 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
 217 * actually.
 218 */
 219struct super_block *freeze_bdev(struct block_device *bdev)
 220{
 221        struct super_block *sb;
 222        int error = 0;
 223
 224        mutex_lock(&bdev->bd_fsfreeze_mutex);
 225        if (++bdev->bd_fsfreeze_count > 1) {
 226                /*
 227                 * We don't even need to grab a reference - the first call
 228                 * to freeze_bdev grab an active reference and only the last
 229                 * thaw_bdev drops it.
 230                 */
 231                sb = get_super(bdev);
 232                drop_super(sb);
 233                mutex_unlock(&bdev->bd_fsfreeze_mutex);
 234                return sb;
 235        }
 236
 237        sb = get_active_super(bdev);
 238        if (!sb)
 239                goto out;
 240        error = freeze_super(sb);
 241        if (error) {
 242                deactivate_super(sb);
 243                bdev->bd_fsfreeze_count--;
 244                mutex_unlock(&bdev->bd_fsfreeze_mutex);
 245                return ERR_PTR(error);
 246        }
 247        deactivate_super(sb);
 248 out:
 249        sync_blockdev(bdev);
 250        mutex_unlock(&bdev->bd_fsfreeze_mutex);
 251        return sb;      /* thaw_bdev releases s->s_umount */
 252}
 253EXPORT_SYMBOL(freeze_bdev);
 254
 255/**
 256 * thaw_bdev  -- unlock filesystem
 257 * @bdev:       blockdevice to unlock
 258 * @sb:         associated superblock
 259 *
 260 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
 261 */
 262int thaw_bdev(struct block_device *bdev, struct super_block *sb)
 263{
 264        int error = -EINVAL;
 265
 266        mutex_lock(&bdev->bd_fsfreeze_mutex);
 267        if (!bdev->bd_fsfreeze_count)
 268                goto out;
 269
 270        error = 0;
 271        if (--bdev->bd_fsfreeze_count > 0)
 272                goto out;
 273
 274        if (!sb)
 275                goto out;
 276
 277        error = thaw_super(sb);
 278        if (error) {
 279                bdev->bd_fsfreeze_count++;
 280                mutex_unlock(&bdev->bd_fsfreeze_mutex);
 281                return error;
 282        }
 283out:
 284        mutex_unlock(&bdev->bd_fsfreeze_mutex);
 285        return 0;
 286}
 287EXPORT_SYMBOL(thaw_bdev);
 288
 289static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
 290{
 291        return block_write_full_page(page, blkdev_get_block, wbc);
 292}
 293
 294static int blkdev_readpage(struct file * file, struct page * page)
 295{
 296        return block_read_full_page(page, blkdev_get_block);
 297}
 298
 299static int blkdev_write_begin(struct file *file, struct address_space *mapping,
 300                        loff_t pos, unsigned len, unsigned flags,
 301                        struct page **pagep, void **fsdata)
 302{
 303        return block_write_begin(mapping, pos, len, flags, pagep,
 304                                 blkdev_get_block);
 305}
 306
 307static int blkdev_write_end(struct file *file, struct address_space *mapping,
 308                        loff_t pos, unsigned len, unsigned copied,
 309                        struct page *page, void *fsdata)
 310{
 311        int ret;
 312        ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
 313
 314        unlock_page(page);
 315        page_cache_release(page);
 316
 317        return ret;
 318}
 319
 320/*
 321 * private llseek:
 322 * for a block special file file_inode(file)->i_size is zero
 323 * so we compute the size by hand (just as in block_read/write above)
 324 */
 325static loff_t block_llseek(struct file *file, loff_t offset, int whence)
 326{
 327        struct inode *bd_inode = file->f_mapping->host;
 328        loff_t size;
 329        loff_t retval;
 330
 331        mutex_lock(&bd_inode->i_mutex);
 332        size = i_size_read(bd_inode);
 333
 334        retval = -EINVAL;
 335        switch (whence) {
 336                case SEEK_END:
 337                        offset += size;
 338                        break;
 339                case SEEK_CUR:
 340                        offset += file->f_pos;
 341                case SEEK_SET:
 342                        break;
 343                default:
 344                        goto out;
 345        }
 346        if (offset >= 0 && offset <= size) {
 347                if (offset != file->f_pos) {
 348                        file->f_pos = offset;
 349                }
 350                retval = offset;
 351        }
 352out:
 353        mutex_unlock(&bd_inode->i_mutex);
 354        return retval;
 355}
 356        
 357int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
 358{
 359        struct inode *bd_inode = filp->f_mapping->host;
 360        struct block_device *bdev = I_BDEV(bd_inode);
 361        int error;
 362        
 363        error = filemap_write_and_wait_range(filp->f_mapping, start, end);
 364        if (error)
 365                return error;
 366
 367        /*
 368         * There is no need to serialise calls to blkdev_issue_flush with
 369         * i_mutex and doing so causes performance issues with concurrent
 370         * O_SYNC writers to a block device.
 371         */
 372        error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
 373        if (error == -EOPNOTSUPP)
 374                error = 0;
 375
 376        return error;
 377}
 378EXPORT_SYMBOL(blkdev_fsync);
 379
 380/*
 381 * pseudo-fs
 382 */
 383
 384static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
 385static struct kmem_cache * bdev_cachep __read_mostly;
 386
 387static struct inode *bdev_alloc_inode(struct super_block *sb)
 388{
 389        struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
 390        if (!ei)
 391                return NULL;
 392        return &ei->vfs_inode;
 393}
 394
 395static void bdev_i_callback(struct rcu_head *head)
 396{
 397        struct inode *inode = container_of(head, struct inode, i_rcu);
 398        struct bdev_inode *bdi = BDEV_I(inode);
 399
 400        kmem_cache_free(bdev_cachep, bdi);
 401}
 402
 403static void bdev_destroy_inode(struct inode *inode)
 404{
 405        call_rcu(&inode->i_rcu, bdev_i_callback);
 406}
 407
 408static void init_once(void *foo)
 409{
 410        struct bdev_inode *ei = (struct bdev_inode *) foo;
 411        struct block_device *bdev = &ei->bdev;
 412
 413        memset(bdev, 0, sizeof(*bdev));
 414        mutex_init(&bdev->bd_mutex);
 415        INIT_LIST_HEAD(&bdev->bd_inodes);
 416        INIT_LIST_HEAD(&bdev->bd_list);
 417#ifdef CONFIG_SYSFS
 418        INIT_LIST_HEAD(&bdev->bd_holder_disks);
 419#endif
 420        inode_init_once(&ei->vfs_inode);
 421        /* Initialize mutex for freeze. */
 422        mutex_init(&bdev->bd_fsfreeze_mutex);
 423}
 424
 425static inline void __bd_forget(struct inode *inode)
 426{
 427        list_del_init(&inode->i_devices);
 428        inode->i_bdev = NULL;
 429        inode->i_mapping = &inode->i_data;
 430}
 431
 432static void bdev_evict_inode(struct inode *inode)
 433{
 434        struct block_device *bdev = &BDEV_I(inode)->bdev;
 435        struct list_head *p;
 436        truncate_inode_pages(&inode->i_data, 0);
 437        invalidate_inode_buffers(inode); /* is it needed here? */
 438        clear_inode(inode);
 439        spin_lock(&bdev_lock);
 440        while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
 441                __bd_forget(list_entry(p, struct inode, i_devices));
 442        }
 443        list_del_init(&bdev->bd_list);
 444        spin_unlock(&bdev_lock);
 445}
 446
 447static const struct super_operations bdev_sops = {
 448        .statfs = simple_statfs,
 449        .alloc_inode = bdev_alloc_inode,
 450        .destroy_inode = bdev_destroy_inode,
 451        .drop_inode = generic_delete_inode,
 452        .evict_inode = bdev_evict_inode,
 453};
 454
 455static struct dentry *bd_mount(struct file_system_type *fs_type,
 456        int flags, const char *dev_name, void *data)
 457{
 458        return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
 459}
 460
 461static struct file_system_type bd_type = {
 462        .name           = "bdev",
 463        .mount          = bd_mount,
 464        .kill_sb        = kill_anon_super,
 465};
 466
 467static struct super_block *blockdev_superblock __read_mostly;
 468
 469void __init bdev_cache_init(void)
 470{
 471        int err;
 472        static struct vfsmount *bd_mnt;
 473
 474        bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
 475                        0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
 476                                SLAB_MEM_SPREAD|SLAB_PANIC),
 477                        init_once);
 478        err = register_filesystem(&bd_type);
 479        if (err)
 480                panic("Cannot register bdev pseudo-fs");
 481        bd_mnt = kern_mount(&bd_type);
 482        if (IS_ERR(bd_mnt))
 483                panic("Cannot create bdev pseudo-fs");
 484        blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
 485}
 486
 487/*
 488 * Most likely _very_ bad one - but then it's hardly critical for small
 489 * /dev and can be fixed when somebody will need really large one.
 490 * Keep in mind that it will be fed through icache hash function too.
 491 */
 492static inline unsigned long hash(dev_t dev)
 493{
 494        return MAJOR(dev)+MINOR(dev);
 495}
 496
 497static int bdev_test(struct inode *inode, void *data)
 498{
 499        return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
 500}
 501
 502static int bdev_set(struct inode *inode, void *data)
 503{
 504        BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
 505        return 0;
 506}
 507
 508static LIST_HEAD(all_bdevs);
 509
 510struct block_device *bdget(dev_t dev)
 511{
 512        struct block_device *bdev;
 513        struct inode *inode;
 514
 515        inode = iget5_locked(blockdev_superblock, hash(dev),
 516                        bdev_test, bdev_set, &dev);
 517
 518        if (!inode)
 519                return NULL;
 520
 521        bdev = &BDEV_I(inode)->bdev;
 522
 523        if (inode->i_state & I_NEW) {
 524                bdev->bd_contains = NULL;
 525                bdev->bd_super = NULL;
 526                bdev->bd_inode = inode;
 527                bdev->bd_block_size = (1 << inode->i_blkbits);
 528                bdev->bd_part_count = 0;
 529                bdev->bd_invalidated = 0;
 530                inode->i_mode = S_IFBLK;
 531                inode->i_rdev = dev;
 532                inode->i_bdev = bdev;
 533                inode->i_data.a_ops = &def_blk_aops;
 534                mapping_set_gfp_mask(&inode->i_data, GFP_USER);
 535                inode->i_data.backing_dev_info = &default_backing_dev_info;
 536                spin_lock(&bdev_lock);
 537                list_add(&bdev->bd_list, &all_bdevs);
 538                spin_unlock(&bdev_lock);
 539                unlock_new_inode(inode);
 540        }
 541        return bdev;
 542}
 543
 544EXPORT_SYMBOL(bdget);
 545
 546/**
 547 * bdgrab -- Grab a reference to an already referenced block device
 548 * @bdev:       Block device to grab a reference to.
 549 */
 550struct block_device *bdgrab(struct block_device *bdev)
 551{
 552        ihold(bdev->bd_inode);
 553        return bdev;
 554}
 555EXPORT_SYMBOL(bdgrab);
 556
 557long nr_blockdev_pages(void)
 558{
 559        struct block_device *bdev;
 560        long ret = 0;
 561        spin_lock(&bdev_lock);
 562        list_for_each_entry(bdev, &all_bdevs, bd_list) {
 563                ret += bdev->bd_inode->i_mapping->nrpages;
 564        }
 565        spin_unlock(&bdev_lock);
 566        return ret;
 567}
 568
 569void bdput(struct block_device *bdev)
 570{
 571        iput(bdev->bd_inode);
 572}
 573
 574EXPORT_SYMBOL(bdput);
 575 
 576static struct block_device *bd_acquire(struct inode *inode)
 577{
 578        struct block_device *bdev;
 579
 580        spin_lock(&bdev_lock);
 581        bdev = inode->i_bdev;
 582        if (bdev) {
 583                ihold(bdev->bd_inode);
 584                spin_unlock(&bdev_lock);
 585                return bdev;
 586        }
 587        spin_unlock(&bdev_lock);
 588
 589        bdev = bdget(inode->i_rdev);
 590        if (bdev) {
 591                spin_lock(&bdev_lock);
 592                if (!inode->i_bdev) {
 593                        /*
 594                         * We take an additional reference to bd_inode,
 595                         * and it's released in clear_inode() of inode.
 596                         * So, we can access it via ->i_mapping always
 597                         * without igrab().
 598                         */
 599                        ihold(bdev->bd_inode);
 600                        inode->i_bdev = bdev;
 601                        inode->i_mapping = bdev->bd_inode->i_mapping;
 602                        list_add(&inode->i_devices, &bdev->bd_inodes);
 603                }
 604                spin_unlock(&bdev_lock);
 605        }
 606        return bdev;
 607}
 608
 609static inline int sb_is_blkdev_sb(struct super_block *sb)
 610{
 611        return sb == blockdev_superblock;
 612}
 613
 614/* Call when you free inode */
 615
 616void bd_forget(struct inode *inode)
 617{
 618        struct block_device *bdev = NULL;
 619
 620        spin_lock(&bdev_lock);
 621        if (!sb_is_blkdev_sb(inode->i_sb))
 622                bdev = inode->i_bdev;
 623        __bd_forget(inode);
 624        spin_unlock(&bdev_lock);
 625
 626        if (bdev)
 627                iput(bdev->bd_inode);
 628}
 629
 630/**
 631 * bd_may_claim - test whether a block device can be claimed
 632 * @bdev: block device of interest
 633 * @whole: whole block device containing @bdev, may equal @bdev
 634 * @holder: holder trying to claim @bdev
 635 *
 636 * Test whether @bdev can be claimed by @holder.
 637 *
 638 * CONTEXT:
 639 * spin_lock(&bdev_lock).
 640 *
 641 * RETURNS:
 642 * %true if @bdev can be claimed, %false otherwise.
 643 */
 644static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
 645                         void *holder)
 646{
 647        if (bdev->bd_holder == holder)
 648                return true;     /* already a holder */
 649        else if (bdev->bd_holder != NULL)
 650                return false;    /* held by someone else */
 651        else if (bdev->bd_contains == bdev)
 652                return true;     /* is a whole device which isn't held */
 653
 654        else if (whole->bd_holder == bd_may_claim)
 655                return true;     /* is a partition of a device that is being partitioned */
 656        else if (whole->bd_holder != NULL)
 657                return false;    /* is a partition of a held device */
 658        else
 659                return true;     /* is a partition of an un-held device */
 660}
 661
 662/**
 663 * bd_prepare_to_claim - prepare to claim a block device
 664 * @bdev: block device of interest
 665 * @whole: the whole device containing @bdev, may equal @bdev
 666 * @holder: holder trying to claim @bdev
 667 *
 668 * Prepare to claim @bdev.  This function fails if @bdev is already
 669 * claimed by another holder and waits if another claiming is in
 670 * progress.  This function doesn't actually claim.  On successful
 671 * return, the caller has ownership of bd_claiming and bd_holder[s].
 672 *
 673 * CONTEXT:
 674 * spin_lock(&bdev_lock).  Might release bdev_lock, sleep and regrab
 675 * it multiple times.
 676 *
 677 * RETURNS:
 678 * 0 if @bdev can be claimed, -EBUSY otherwise.
 679 */
 680static int bd_prepare_to_claim(struct block_device *bdev,
 681                               struct block_device *whole, void *holder)
 682{
 683retry:
 684        /* if someone else claimed, fail */
 685        if (!bd_may_claim(bdev, whole, holder))
 686                return -EBUSY;
 687
 688        /* if claiming is already in progress, wait for it to finish */
 689        if (whole->bd_claiming) {
 690                wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
 691                DEFINE_WAIT(wait);
 692
 693                prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
 694                spin_unlock(&bdev_lock);
 695                schedule();
 696                finish_wait(wq, &wait);
 697                spin_lock(&bdev_lock);
 698                goto retry;
 699        }
 700
 701        /* yay, all mine */
 702        return 0;
 703}
 704
 705/**
 706 * bd_start_claiming - start claiming a block device
 707 * @bdev: block device of interest
 708 * @holder: holder trying to claim @bdev
 709 *
 710 * @bdev is about to be opened exclusively.  Check @bdev can be opened
 711 * exclusively and mark that an exclusive open is in progress.  Each
 712 * successful call to this function must be matched with a call to
 713 * either bd_finish_claiming() or bd_abort_claiming() (which do not
 714 * fail).
 715 *
 716 * This function is used to gain exclusive access to the block device
 717 * without actually causing other exclusive open attempts to fail. It
 718 * should be used when the open sequence itself requires exclusive
 719 * access but may subsequently fail.
 720 *
 721 * CONTEXT:
 722 * Might sleep.
 723 *
 724 * RETURNS:
 725 * Pointer to the block device containing @bdev on success, ERR_PTR()
 726 * value on failure.
 727 */
 728static struct block_device *bd_start_claiming(struct block_device *bdev,
 729                                              void *holder)
 730{
 731        struct gendisk *disk;
 732        struct block_device *whole;
 733        int partno, err;
 734
 735        might_sleep();
 736
 737        /*
 738         * @bdev might not have been initialized properly yet, look up
 739         * and grab the outer block device the hard way.
 740         */
 741        disk = get_gendisk(bdev->bd_dev, &partno);
 742        if (!disk)
 743                return ERR_PTR(-ENXIO);
 744
 745        /*
 746         * Normally, @bdev should equal what's returned from bdget_disk()
 747         * if partno is 0; however, some drivers (floppy) use multiple
 748         * bdev's for the same physical device and @bdev may be one of the
 749         * aliases.  Keep @bdev if partno is 0.  This means claimer
 750         * tracking is broken for those devices but it has always been that
 751         * way.
 752         */
 753        if (partno)
 754                whole = bdget_disk(disk, 0);
 755        else
 756                whole = bdgrab(bdev);
 757
 758        module_put(disk->fops->owner);
 759        put_disk(disk);
 760        if (!whole)
 761                return ERR_PTR(-ENOMEM);
 762
 763        /* prepare to claim, if successful, mark claiming in progress */
 764        spin_lock(&bdev_lock);
 765
 766        err = bd_prepare_to_claim(bdev, whole, holder);
 767        if (err == 0) {
 768                whole->bd_claiming = holder;
 769                spin_unlock(&bdev_lock);
 770                return whole;
 771        } else {
 772                spin_unlock(&bdev_lock);
 773                bdput(whole);
 774                return ERR_PTR(err);
 775        }
 776}
 777
 778#ifdef CONFIG_SYSFS
 779struct bd_holder_disk {
 780        struct list_head        list;
 781        struct gendisk          *disk;
 782        int                     refcnt;
 783};
 784
 785static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
 786                                                  struct gendisk *disk)
 787{
 788        struct bd_holder_disk *holder;
 789
 790        list_for_each_entry(holder, &bdev->bd_holder_disks, list)
 791                if (holder->disk == disk)
 792                        return holder;
 793        return NULL;
 794}
 795
 796static int add_symlink(struct kobject *from, struct kobject *to)
 797{
 798        return sysfs_create_link(from, to, kobject_name(to));
 799}
 800
 801static void del_symlink(struct kobject *from, struct kobject *to)
 802{
 803        sysfs_remove_link(from, kobject_name(to));
 804}
 805
 806/**
 807 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
 808 * @bdev: the claimed slave bdev
 809 * @disk: the holding disk
 810 *
 811 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
 812 *
 813 * This functions creates the following sysfs symlinks.
 814 *
 815 * - from "slaves" directory of the holder @disk to the claimed @bdev
 816 * - from "holders" directory of the @bdev to the holder @disk
 817 *
 818 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
 819 * passed to bd_link_disk_holder(), then:
 820 *
 821 *   /sys/block/dm-0/slaves/sda --> /sys/block/sda
 822 *   /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
 823 *
 824 * The caller must have claimed @bdev before calling this function and
 825 * ensure that both @bdev and @disk are valid during the creation and
 826 * lifetime of these symlinks.
 827 *
 828 * CONTEXT:
 829 * Might sleep.
 830 *
 831 * RETURNS:
 832 * 0 on success, -errno on failure.
 833 */
 834int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
 835{
 836        struct bd_holder_disk *holder;
 837        int ret = 0;
 838
 839        mutex_lock(&bdev->bd_mutex);
 840
 841        WARN_ON_ONCE(!bdev->bd_holder);
 842
 843        /* FIXME: remove the following once add_disk() handles errors */
 844        if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
 845                goto out_unlock;
 846
 847        holder = bd_find_holder_disk(bdev, disk);
 848        if (holder) {
 849                holder->refcnt++;
 850                goto out_unlock;
 851        }
 852
 853        holder = kzalloc(sizeof(*holder), GFP_KERNEL);
 854        if (!holder) {
 855                ret = -ENOMEM;
 856                goto out_unlock;
 857        }
 858
 859        INIT_LIST_HEAD(&holder->list);
 860        holder->disk = disk;
 861        holder->refcnt = 1;
 862
 863        ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
 864        if (ret)
 865                goto out_free;
 866
 867        ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
 868        if (ret)
 869                goto out_del;
 870        /*
 871         * bdev could be deleted beneath us which would implicitly destroy
 872         * the holder directory.  Hold on to it.
 873         */
 874        kobject_get(bdev->bd_part->holder_dir);
 875
 876        list_add(&holder->list, &bdev->bd_holder_disks);
 877        goto out_unlock;
 878
 879out_del:
 880        del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
 881out_free:
 882        kfree(holder);
 883out_unlock:
 884        mutex_unlock(&bdev->bd_mutex);
 885        return ret;
 886}
 887EXPORT_SYMBOL_GPL(bd_link_disk_holder);
 888
 889/**
 890 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
 891 * @bdev: the calimed slave bdev
 892 * @disk: the holding disk
 893 *
 894 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
 895 *
 896 * CONTEXT:
 897 * Might sleep.
 898 */
 899void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
 900{
 901        struct bd_holder_disk *holder;
 902
 903        mutex_lock(&bdev->bd_mutex);
 904
 905        holder = bd_find_holder_disk(bdev, disk);
 906
 907        if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
 908                del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
 909                del_symlink(bdev->bd_part->holder_dir,
 910                            &disk_to_dev(disk)->kobj);
 911                kobject_put(bdev->bd_part->holder_dir);
 912                list_del_init(&holder->list);
 913                kfree(holder);
 914        }
 915
 916        mutex_unlock(&bdev->bd_mutex);
 917}
 918EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
 919#endif
 920
 921/**
 922 * flush_disk - invalidates all buffer-cache entries on a disk
 923 *
 924 * @bdev:      struct block device to be flushed
 925 * @kill_dirty: flag to guide handling of dirty inodes
 926 *
 927 * Invalidates all buffer-cache entries on a disk. It should be called
 928 * when a disk has been changed -- either by a media change or online
 929 * resize.
 930 */
 931static void flush_disk(struct block_device *bdev, bool kill_dirty)
 932{
 933        if (__invalidate_device(bdev, kill_dirty)) {
 934                char name[BDEVNAME_SIZE] = "";
 935
 936                if (bdev->bd_disk)
 937                        disk_name(bdev->bd_disk, 0, name);
 938                printk(KERN_WARNING "VFS: busy inodes on changed media or "
 939                       "resized disk %s\n", name);
 940        }
 941
 942        if (!bdev->bd_disk)
 943                return;
 944        if (disk_part_scan_enabled(bdev->bd_disk))
 945                bdev->bd_invalidated = 1;
 946}
 947
 948/**
 949 * check_disk_size_change - checks for disk size change and adjusts bdev size.
 950 * @disk: struct gendisk to check
 951 * @bdev: struct bdev to adjust.
 952 *
 953 * This routine checks to see if the bdev size does not match the disk size
 954 * and adjusts it if it differs.
 955 */
 956void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
 957{
 958        loff_t disk_size, bdev_size;
 959
 960        disk_size = (loff_t)get_capacity(disk) << 9;
 961        bdev_size = i_size_read(bdev->bd_inode);
 962        if (disk_size != bdev_size) {
 963                char name[BDEVNAME_SIZE];
 964
 965                disk_name(disk, 0, name);
 966                printk(KERN_INFO
 967                       "%s: detected capacity change from %lld to %lld\n",
 968                       name, bdev_size, disk_size);
 969                i_size_write(bdev->bd_inode, disk_size);
 970                flush_disk(bdev, false);
 971        }
 972}
 973EXPORT_SYMBOL(check_disk_size_change);
 974
 975/**
 976 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
 977 * @disk: struct gendisk to be revalidated
 978 *
 979 * This routine is a wrapper for lower-level driver's revalidate_disk
 980 * call-backs.  It is used to do common pre and post operations needed
 981 * for all revalidate_disk operations.
 982 */
 983int revalidate_disk(struct gendisk *disk)
 984{
 985        struct block_device *bdev;
 986        int ret = 0;
 987
 988        if (disk->fops->revalidate_disk)
 989                ret = disk->fops->revalidate_disk(disk);
 990
 991        bdev = bdget_disk(disk, 0);
 992        if (!bdev)
 993                return ret;
 994
 995        mutex_lock(&bdev->bd_mutex);
 996        check_disk_size_change(disk, bdev);
 997        bdev->bd_invalidated = 0;
 998        mutex_unlock(&bdev->bd_mutex);
 999        bdput(bdev);
1000        return ret;
1001}
1002EXPORT_SYMBOL(revalidate_disk);
1003
1004/*
1005 * This routine checks whether a removable media has been changed,
1006 * and invalidates all buffer-cache-entries in that case. This
1007 * is a relatively slow routine, so we have to try to minimize using
1008 * it. Thus it is called only upon a 'mount' or 'open'. This
1009 * is the best way of combining speed and utility, I think.
1010 * People changing diskettes in the middle of an operation deserve
1011 * to lose :-)
1012 */
1013int check_disk_change(struct block_device *bdev)
1014{
1015        struct gendisk *disk = bdev->bd_disk;
1016        const struct block_device_operations *bdops = disk->fops;
1017        unsigned int events;
1018
1019        events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1020                                   DISK_EVENT_EJECT_REQUEST);
1021        if (!(events & DISK_EVENT_MEDIA_CHANGE))
1022                return 0;
1023
1024        flush_disk(bdev, true);
1025        if (bdops->revalidate_disk)
1026                bdops->revalidate_disk(bdev->bd_disk);
1027        return 1;
1028}
1029
1030EXPORT_SYMBOL(check_disk_change);
1031
1032void bd_set_size(struct block_device *bdev, loff_t size)
1033{
1034        unsigned bsize = bdev_logical_block_size(bdev);
1035
1036        mutex_lock(&bdev->bd_inode->i_mutex);
1037        i_size_write(bdev->bd_inode, size);
1038        mutex_unlock(&bdev->bd_inode->i_mutex);
1039        while (bsize < PAGE_CACHE_SIZE) {
1040                if (size & bsize)
1041                        break;
1042                bsize <<= 1;
1043        }
1044        bdev->bd_block_size = bsize;
1045        bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1046}
1047EXPORT_SYMBOL(bd_set_size);
1048
1049static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1050
1051/*
1052 * bd_mutex locking:
1053 *
1054 *  mutex_lock(part->bd_mutex)
1055 *    mutex_lock_nested(whole->bd_mutex, 1)
1056 */
1057
1058static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1059{
1060        struct gendisk *disk;
1061        struct module *owner;
1062        int ret;
1063        int partno;
1064        int perm = 0;
1065
1066        if (mode & FMODE_READ)
1067                perm |= MAY_READ;
1068        if (mode & FMODE_WRITE)
1069                perm |= MAY_WRITE;
1070        /*
1071         * hooks: /n/, see "layering violations".
1072         */
1073        if (!for_part) {
1074                ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1075                if (ret != 0) {
1076                        bdput(bdev);
1077                        return ret;
1078                }
1079        }
1080
1081 restart:
1082
1083        ret = -ENXIO;
1084        disk = get_gendisk(bdev->bd_dev, &partno);
1085        if (!disk)
1086                goto out;
1087        owner = disk->fops->owner;
1088
1089        disk_block_events(disk);
1090        mutex_lock_nested(&bdev->bd_mutex, for_part);
1091        if (!bdev->bd_openers) {
1092                bdev->bd_disk = disk;
1093                bdev->bd_queue = disk->queue;
1094                bdev->bd_contains = bdev;
1095                if (!partno) {
1096                        struct backing_dev_info *bdi;
1097
1098                        ret = -ENXIO;
1099                        bdev->bd_part = disk_get_part(disk, partno);
1100                        if (!bdev->bd_part)
1101                                goto out_clear;
1102
1103                        ret = 0;
1104                        if (disk->fops->open) {
1105                                ret = disk->fops->open(bdev, mode);
1106                                if (ret == -ERESTARTSYS) {
1107                                        /* Lost a race with 'disk' being
1108                                         * deleted, try again.
1109                                         * See md.c
1110                                         */
1111                                        disk_put_part(bdev->bd_part);
1112                                        bdev->bd_part = NULL;
1113                                        bdev->bd_disk = NULL;
1114                                        bdev->bd_queue = NULL;
1115                                        mutex_unlock(&bdev->bd_mutex);
1116                                        disk_unblock_events(disk);
1117                                        put_disk(disk);
1118                                        module_put(owner);
1119                                        goto restart;
1120                                }
1121                        }
1122
1123                        if (!ret) {
1124                                bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1125                                bdi = blk_get_backing_dev_info(bdev);
1126                                if (bdi == NULL)
1127                                        bdi = &default_backing_dev_info;
1128                                bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1129                        }
1130
1131                        /*
1132                         * If the device is invalidated, rescan partition
1133                         * if open succeeded or failed with -ENOMEDIUM.
1134                         * The latter is necessary to prevent ghost
1135                         * partitions on a removed medium.
1136                         */
1137                        if (bdev->bd_invalidated) {
1138                                if (!ret)
1139                                        rescan_partitions(disk, bdev);
1140                                else if (ret == -ENOMEDIUM)
1141                                        invalidate_partitions(disk, bdev);
1142                        }
1143                        if (ret)
1144                                goto out_clear;
1145                } else {
1146                        struct block_device *whole;
1147                        whole = bdget_disk(disk, 0);
1148                        ret = -ENOMEM;
1149                        if (!whole)
1150                                goto out_clear;
1151                        BUG_ON(for_part);
1152                        ret = __blkdev_get(whole, mode, 1);
1153                        if (ret)
1154                                goto out_clear;
1155                        bdev->bd_contains = whole;
1156                        bdev_inode_switch_bdi(bdev->bd_inode,
1157                                whole->bd_inode->i_data.backing_dev_info);
1158                        bdev->bd_part = disk_get_part(disk, partno);
1159                        if (!(disk->flags & GENHD_FL_UP) ||
1160                            !bdev->bd_part || !bdev->bd_part->nr_sects) {
1161                                ret = -ENXIO;
1162                                goto out_clear;
1163                        }
1164                        bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1165                }
1166        } else {
1167                if (bdev->bd_contains == bdev) {
1168                        ret = 0;
1169                        if (bdev->bd_disk->fops->open)
1170                                ret = bdev->bd_disk->fops->open(bdev, mode);
1171                        /* the same as first opener case, read comment there */
1172                        if (bdev->bd_invalidated) {
1173                                if (!ret)
1174                                        rescan_partitions(bdev->bd_disk, bdev);
1175                                else if (ret == -ENOMEDIUM)
1176                                        invalidate_partitions(bdev->bd_disk, bdev);
1177                        }
1178                        if (ret)
1179                                goto out_unlock_bdev;
1180                }
1181                /* only one opener holds refs to the module and disk */
1182                put_disk(disk);
1183                module_put(owner);
1184        }
1185        bdev->bd_openers++;
1186        if (for_part)
1187                bdev->bd_part_count++;
1188        mutex_unlock(&bdev->bd_mutex);
1189        disk_unblock_events(disk);
1190        return 0;
1191
1192 out_clear:
1193        disk_put_part(bdev->bd_part);
1194        bdev->bd_disk = NULL;
1195        bdev->bd_part = NULL;
1196        bdev->bd_queue = NULL;
1197        bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1198        if (bdev != bdev->bd_contains)
1199                __blkdev_put(bdev->bd_contains, mode, 1);
1200        bdev->bd_contains = NULL;
1201 out_unlock_bdev:
1202        mutex_unlock(&bdev->bd_mutex);
1203        disk_unblock_events(disk);
1204        put_disk(disk);
1205        module_put(owner);
1206 out:
1207        bdput(bdev);
1208
1209        return ret;
1210}
1211
1212/**
1213 * blkdev_get - open a block device
1214 * @bdev: block_device to open
1215 * @mode: FMODE_* mask
1216 * @holder: exclusive holder identifier
1217 *
1218 * Open @bdev with @mode.  If @mode includes %FMODE_EXCL, @bdev is
1219 * open with exclusive access.  Specifying %FMODE_EXCL with %NULL
1220 * @holder is invalid.  Exclusive opens may nest for the same @holder.
1221 *
1222 * On success, the reference count of @bdev is unchanged.  On failure,
1223 * @bdev is put.
1224 *
1225 * CONTEXT:
1226 * Might sleep.
1227 *
1228 * RETURNS:
1229 * 0 on success, -errno on failure.
1230 */
1231int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1232{
1233        struct block_device *whole = NULL;
1234        int res;
1235
1236        WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1237
1238        if ((mode & FMODE_EXCL) && holder) {
1239                whole = bd_start_claiming(bdev, holder);
1240                if (IS_ERR(whole)) {
1241                        bdput(bdev);
1242                        return PTR_ERR(whole);
1243                }
1244        }
1245
1246        res = __blkdev_get(bdev, mode, 0);
1247
1248        if (whole) {
1249                struct gendisk *disk = whole->bd_disk;
1250
1251                /* finish claiming */
1252                mutex_lock(&bdev->bd_mutex);
1253                spin_lock(&bdev_lock);
1254
1255                if (!res) {
1256                        BUG_ON(!bd_may_claim(bdev, whole, holder));
1257                        /*
1258                         * Note that for a whole device bd_holders
1259                         * will be incremented twice, and bd_holder
1260                         * will be set to bd_may_claim before being
1261                         * set to holder
1262                         */
1263                        whole->bd_holders++;
1264                        whole->bd_holder = bd_may_claim;
1265                        bdev->bd_holders++;
1266                        bdev->bd_holder = holder;
1267                }
1268
1269                /* tell others that we're done */
1270                BUG_ON(whole->bd_claiming != holder);
1271                whole->bd_claiming = NULL;
1272                wake_up_bit(&whole->bd_claiming, 0);
1273
1274                spin_unlock(&bdev_lock);
1275
1276                /*
1277                 * Block event polling for write claims if requested.  Any
1278                 * write holder makes the write_holder state stick until
1279                 * all are released.  This is good enough and tracking
1280                 * individual writeable reference is too fragile given the
1281                 * way @mode is used in blkdev_get/put().
1282                 */
1283                if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1284                    (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1285                        bdev->bd_write_holder = true;
1286                        disk_block_events(disk);
1287                }
1288
1289                mutex_unlock(&bdev->bd_mutex);
1290                bdput(whole);
1291        }
1292
1293        return res;
1294}
1295EXPORT_SYMBOL(blkdev_get);
1296
1297/**
1298 * blkdev_get_by_path - open a block device by name
1299 * @path: path to the block device to open
1300 * @mode: FMODE_* mask
1301 * @holder: exclusive holder identifier
1302 *
1303 * Open the blockdevice described by the device file at @path.  @mode
1304 * and @holder are identical to blkdev_get().
1305 *
1306 * On success, the returned block_device has reference count of one.
1307 *
1308 * CONTEXT:
1309 * Might sleep.
1310 *
1311 * RETURNS:
1312 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1313 */
1314struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1315                                        void *holder)
1316{
1317        struct block_device *bdev;
1318        int err;
1319
1320        bdev = lookup_bdev(path);
1321        if (IS_ERR(bdev))
1322                return bdev;
1323
1324        err = blkdev_get(bdev, mode, holder);
1325        if (err)
1326                return ERR_PTR(err);
1327
1328        if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1329                blkdev_put(bdev, mode);
1330                return ERR_PTR(-EACCES);
1331        }
1332
1333        return bdev;
1334}
1335EXPORT_SYMBOL(blkdev_get_by_path);
1336
1337/**
1338 * blkdev_get_by_dev - open a block device by device number
1339 * @dev: device number of block device to open
1340 * @mode: FMODE_* mask
1341 * @holder: exclusive holder identifier
1342 *
1343 * Open the blockdevice described by device number @dev.  @mode and
1344 * @holder are identical to blkdev_get().
1345 *
1346 * Use it ONLY if you really do not have anything better - i.e. when
1347 * you are behind a truly sucky interface and all you are given is a
1348 * device number.  _Never_ to be used for internal purposes.  If you
1349 * ever need it - reconsider your API.
1350 *
1351 * On success, the returned block_device has reference count of one.
1352 *
1353 * CONTEXT:
1354 * Might sleep.
1355 *
1356 * RETURNS:
1357 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1358 */
1359struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1360{
1361        struct block_device *bdev;
1362        int err;
1363
1364        bdev = bdget(dev);
1365        if (!bdev)
1366                return ERR_PTR(-ENOMEM);
1367
1368        err = blkdev_get(bdev, mode, holder);
1369        if (err)
1370                return ERR_PTR(err);
1371
1372        return bdev;
1373}
1374EXPORT_SYMBOL(blkdev_get_by_dev);
1375
1376static int blkdev_open(struct inode * inode, struct file * filp)
1377{
1378        struct block_device *bdev;
1379
1380        /*
1381         * Preserve backwards compatibility and allow large file access
1382         * even if userspace doesn't ask for it explicitly. Some mkfs
1383         * binary needs it. We might want to drop this workaround
1384         * during an unstable branch.
1385         */
1386        filp->f_flags |= O_LARGEFILE;
1387
1388        if (filp->f_flags & O_NDELAY)
1389                filp->f_mode |= FMODE_NDELAY;
1390        if (filp->f_flags & O_EXCL)
1391                filp->f_mode |= FMODE_EXCL;
1392        if ((filp->f_flags & O_ACCMODE) == 3)
1393                filp->f_mode |= FMODE_WRITE_IOCTL;
1394
1395        bdev = bd_acquire(inode);
1396        if (bdev == NULL)
1397                return -ENOMEM;
1398
1399        filp->f_mapping = bdev->bd_inode->i_mapping;
1400
1401        return blkdev_get(bdev, filp->f_mode, filp);
1402}
1403
1404static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1405{
1406        struct gendisk *disk = bdev->bd_disk;
1407        struct block_device *victim = NULL;
1408
1409        mutex_lock_nested(&bdev->bd_mutex, for_part);
1410        if (for_part)
1411                bdev->bd_part_count--;
1412
1413        if (!--bdev->bd_openers) {
1414                WARN_ON_ONCE(bdev->bd_holders);
1415                sync_blockdev(bdev);
1416                kill_bdev(bdev);
1417                /* ->release can cause the old bdi to disappear,
1418                 * so must switch it out first
1419                 */
1420                bdev_inode_switch_bdi(bdev->bd_inode,
1421                                        &default_backing_dev_info);
1422        }
1423        if (bdev->bd_contains == bdev) {
1424                if (disk->fops->release)
1425                        disk->fops->release(disk, mode);
1426        }
1427        if (!bdev->bd_openers) {
1428                struct module *owner = disk->fops->owner;
1429
1430                disk_put_part(bdev->bd_part);
1431                bdev->bd_part = NULL;
1432                bdev->bd_disk = NULL;
1433                if (bdev != bdev->bd_contains)
1434                        victim = bdev->bd_contains;
1435                bdev->bd_contains = NULL;
1436
1437                put_disk(disk);
1438                module_put(owner);
1439        }
1440        mutex_unlock(&bdev->bd_mutex);
1441        bdput(bdev);
1442        if (victim)
1443                __blkdev_put(victim, mode, 1);
1444}
1445
1446void blkdev_put(struct block_device *bdev, fmode_t mode)
1447{
1448        mutex_lock(&bdev->bd_mutex);
1449
1450        if (mode & FMODE_EXCL) {
1451                bool bdev_free;
1452
1453                /*
1454                 * Release a claim on the device.  The holder fields
1455                 * are protected with bdev_lock.  bd_mutex is to
1456                 * synchronize disk_holder unlinking.
1457                 */
1458                spin_lock(&bdev_lock);
1459
1460                WARN_ON_ONCE(--bdev->bd_holders < 0);
1461                WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1462
1463                /* bd_contains might point to self, check in a separate step */
1464                if ((bdev_free = !bdev->bd_holders))
1465                        bdev->bd_holder = NULL;
1466                if (!bdev->bd_contains->bd_holders)
1467                        bdev->bd_contains->bd_holder = NULL;
1468
1469                spin_unlock(&bdev_lock);
1470
1471                /*
1472                 * If this was the last claim, remove holder link and
1473                 * unblock evpoll if it was a write holder.
1474                 */
1475                if (bdev_free && bdev->bd_write_holder) {
1476                        disk_unblock_events(bdev->bd_disk);
1477                        bdev->bd_write_holder = false;
1478                }
1479        }
1480
1481        /*
1482         * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1483         * event.  This is to ensure detection of media removal commanded
1484         * from userland - e.g. eject(1).
1485         */
1486        disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1487
1488        mutex_unlock(&bdev->bd_mutex);
1489
1490        __blkdev_put(bdev, mode, 0);
1491}
1492EXPORT_SYMBOL(blkdev_put);
1493
1494static int blkdev_close(struct inode * inode, struct file * filp)
1495{
1496        struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1497        blkdev_put(bdev, filp->f_mode);
1498        return 0;
1499}
1500
1501static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1502{
1503        struct block_device *bdev = I_BDEV(file->f_mapping->host);
1504        fmode_t mode = file->f_mode;
1505
1506        /*
1507         * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1508         * to updated it before every ioctl.
1509         */
1510        if (file->f_flags & O_NDELAY)
1511                mode |= FMODE_NDELAY;
1512        else
1513                mode &= ~FMODE_NDELAY;
1514
1515        return blkdev_ioctl(bdev, mode, cmd, arg);
1516}
1517
1518/*
1519 * Write data to the block device.  Only intended for the block device itself
1520 * and the raw driver which basically is a fake block device.
1521 *
1522 * Does not take i_mutex for the write and thus is not for general purpose
1523 * use.
1524 */
1525ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1526                         unsigned long nr_segs, loff_t pos)
1527{
1528        struct file *file = iocb->ki_filp;
1529        struct blk_plug plug;
1530        ssize_t ret;
1531
1532        BUG_ON(iocb->ki_pos != pos);
1533
1534        blk_start_plug(&plug);
1535        ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1536        if (ret > 0 || ret == -EIOCBQUEUED) {
1537                ssize_t err;
1538
1539                err = generic_write_sync(file, pos, ret);
1540                if (err < 0 && ret > 0)
1541                        ret = err;
1542        }
1543        blk_finish_plug(&plug);
1544        return ret;
1545}
1546EXPORT_SYMBOL_GPL(blkdev_aio_write);
1547
1548static ssize_t blkdev_aio_read(struct kiocb *iocb, const struct iovec *iov,
1549                         unsigned long nr_segs, loff_t pos)
1550{
1551        struct file *file = iocb->ki_filp;
1552        struct inode *bd_inode = file->f_mapping->host;
1553        loff_t size = i_size_read(bd_inode);
1554
1555        if (pos >= size)
1556                return 0;
1557
1558        size -= pos;
1559        if (size < iocb->ki_left)
1560                nr_segs = iov_shorten((struct iovec *)iov, nr_segs, size);
1561        return generic_file_aio_read(iocb, iov, nr_segs, pos);
1562}
1563
1564/*
1565 * Try to release a page associated with block device when the system
1566 * is under memory pressure.
1567 */
1568static int blkdev_releasepage(struct page *page, gfp_t wait)
1569{
1570        struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1571
1572        if (super && super->s_op->bdev_try_to_free_page)
1573                return super->s_op->bdev_try_to_free_page(super, page, wait);
1574
1575        return try_to_free_buffers(page);
1576}
1577
1578static const struct address_space_operations def_blk_aops = {
1579        .readpage       = blkdev_readpage,
1580        .writepage      = blkdev_writepage,
1581        .write_begin    = blkdev_write_begin,
1582        .write_end      = blkdev_write_end,
1583        .writepages     = generic_writepages,
1584        .releasepage    = blkdev_releasepage,
1585        .direct_IO      = blkdev_direct_IO,
1586};
1587
1588const struct file_operations def_blk_fops = {
1589        .open           = blkdev_open,
1590        .release        = blkdev_close,
1591        .llseek         = block_llseek,
1592        .read           = do_sync_read,
1593        .write          = do_sync_write,
1594        .aio_read       = blkdev_aio_read,
1595        .aio_write      = blkdev_aio_write,
1596        .mmap           = generic_file_mmap,
1597        .fsync          = blkdev_fsync,
1598        .unlocked_ioctl = block_ioctl,
1599#ifdef CONFIG_COMPAT
1600        .compat_ioctl   = compat_blkdev_ioctl,
1601#endif
1602        .splice_read    = generic_file_splice_read,
1603        .splice_write   = generic_file_splice_write,
1604};
1605
1606int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1607{
1608        int res;
1609        mm_segment_t old_fs = get_fs();
1610        set_fs(KERNEL_DS);
1611        res = blkdev_ioctl(bdev, 0, cmd, arg);
1612        set_fs(old_fs);
1613        return res;
1614}
1615
1616EXPORT_SYMBOL(ioctl_by_bdev);
1617
1618/**
1619 * lookup_bdev  - lookup a struct block_device by name
1620 * @pathname:   special file representing the block device
1621 *
1622 * Get a reference to the blockdevice at @pathname in the current
1623 * namespace if possible and return it.  Return ERR_PTR(error)
1624 * otherwise.
1625 */
1626struct block_device *lookup_bdev(const char *pathname)
1627{
1628        struct block_device *bdev;
1629        struct inode *inode;
1630        struct path path;
1631        int error;
1632
1633        if (!pathname || !*pathname)
1634                return ERR_PTR(-EINVAL);
1635
1636        error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1637        if (error)
1638                return ERR_PTR(error);
1639
1640        inode = path.dentry->d_inode;
1641        error = -ENOTBLK;
1642        if (!S_ISBLK(inode->i_mode))
1643                goto fail;
1644        error = -EACCES;
1645        if (path.mnt->mnt_flags & MNT_NODEV)
1646                goto fail;
1647        error = -ENOMEM;
1648        bdev = bd_acquire(inode);
1649        if (!bdev)
1650                goto fail;
1651out:
1652        path_put(&path);
1653        return bdev;
1654fail:
1655        bdev = ERR_PTR(error);
1656        goto out;
1657}
1658EXPORT_SYMBOL(lookup_bdev);
1659
1660int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1661{
1662        struct super_block *sb = get_super(bdev);
1663        int res = 0;
1664
1665        if (sb) {
1666                /*
1667                 * no need to lock the super, get_super holds the
1668                 * read mutex so the filesystem cannot go away
1669                 * under us (->put_super runs with the write lock
1670                 * hold).
1671                 */
1672                shrink_dcache_sb(sb);
1673                res = invalidate_inodes(sb, kill_dirty);
1674                drop_super(sb);
1675        }
1676        invalidate_bdev(bdev);
1677        return res;
1678}
1679EXPORT_SYMBOL(__invalidate_device);
1680
1681void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1682{
1683        struct inode *inode, *old_inode = NULL;
1684
1685        spin_lock(&inode_sb_list_lock);
1686        list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1687                struct address_space *mapping = inode->i_mapping;
1688
1689                spin_lock(&inode->i_lock);
1690                if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1691                    mapping->nrpages == 0) {
1692                        spin_unlock(&inode->i_lock);
1693                        continue;
1694                }
1695                __iget(inode);
1696                spin_unlock(&inode->i_lock);
1697                spin_unlock(&inode_sb_list_lock);
1698                /*
1699                 * We hold a reference to 'inode' so it couldn't have been
1700                 * removed from s_inodes list while we dropped the
1701                 * inode_sb_list_lock.  We cannot iput the inode now as we can
1702                 * be holding the last reference and we cannot iput it under
1703                 * inode_sb_list_lock. So we keep the reference and iput it
1704                 * later.
1705                 */
1706                iput(old_inode);
1707                old_inode = inode;
1708
1709                func(I_BDEV(inode), arg);
1710
1711                spin_lock(&inode_sb_list_lock);
1712        }
1713        spin_unlock(&inode_sb_list_lock);
1714        iput(old_inode);
1715}
1716
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.